Sensitivity to Endocrine Therapy Index Predicts Benefit from Weekly Adjuvant Paclitaxel for Hormone Receptor–Positive Breast Cancer in the GEICAM/9906 Trial

Abstract

Purpose:

To independently validate that low endocrine transcriptional activity measured by the sensitivity to endocrine therapy (SET_ER/PR) index in hormone receptor–positive (HR⁺) breast cancer predicts benefit from dose-dense paclitaxel chemotherapy within a second prospective–retrospective biomarker study.

Experimental Design:

We conducted a blinded, prospective–retrospective biomarker analysis within the GEICAM/9906 trial (NCT00129922), which compared adjuvant 5-fluorouracil, epirubicin, and cyclophosphamide (FEC) followed by weekly paclitaxel (P) versus six cycles of FEC in lymph node–positive breast cancer. The SET_ER/PR index was measured in all available HR⁺/HER2⁻ tumor RNA samples using a prespecified cut point (<0.75). The primary endpoint was the distant recurrence–free interval (DRFI); secondary endpoints were overall survival (OS) and breast cancer–specific survival (BCSS).

Results:

Of 647 HR⁺/HER2− tumors, 567 (87.6%) passed assay quality control (279 FEC + P; 288 FEC). A low SET_ER/PR index was identified in 92 tumors (16.2%). There was a significant interaction between SET_ER/PR status and treatment on DRFI (P = 0.046). Among patients with a low SET_ER/PR index, FEC + P significantly improved DRFI [hazard ratio (HR), 0.46; 95% confidence interval (CI), 0.22–0.95; P = 0.035], with similar results after adjustment (HR, 0.48; 95% CI, 0.24–1; P = 0.049). No treatment benefit was observed for SET_ER/PR ≥0.75 (HR, 1.02; 95% CI, 0.70–1.47; P = 0.931). Differences in OS and BCSS did not reach significance.

Conclusions:

Low endocrine transcriptional activity predicts benefit from adding weekly paclitaxel to anthracycline-based adjuvant chemotherapy in HR⁺/HER2⁻ breast cancer. These findings independently validate the SET_ER/PR index as a predictive biomarker for paclitaxel-based chemotherapy and support its potential role in guiding regimen selection.

Translational Relevance.

This prospective–retrospective study from a second randomized clinical trial provides independent confirmation that patients whose hormone receptor–positive (HR⁺) breast cancer has low endocrine transcriptional activity [i.e., sensitivity to endocrine therapy (SET_ER/PR) index below 0.75] benefit from adjuvant chemotherapy that includes a dose-dense paclitaxel treatment schedule, whereas those with higher endocrine transcriptional activity do not benefit. Notably, the SET_ER/PR index is the first genomic assay to predict survival benefit from a contemporary taxane-based chemotherapy regimen for patients with HR⁺ breast cancer. The results from this current study establish clinical utility for a new generation of tests that can predict which chemotherapy regimen would offer survival benefit for an individual patient, thereby contributing to personalized and improved adjuvant treatment selection for these patients.

Introduction

The recent meta-analysis by the Early Breast Cancer Trials Collaborative Group (EBCTCG) demonstrated that improvements in adjuvant chemotherapy from the addition of taxanes have achieved similar relative benefits in hormone receptor–positive (HR⁺) and HR⁻ breast cancer, with no difference by menopausal status (1–3). The general characteristics associated with more effective chemotherapy were the inclusion of an anthracycline with a taxane, a higher cumulative dose, more cycles of treatment, and dose-dense treatment schedules (2, 3). These EBCTCG findings might seem at odds with recent trends in chemotherapy use for patients with HR⁺ breast cancer that aim to deintensify treatment for some patients, including the exclusion of anthracyclines and fewer treatment cycles, and they did not address the use of genomic test results and menopausal status to select which patients with HR⁺ cancer to recommend adjuvant chemotherapy (3–5). However, they do underline the need for predictive biomarkers that would help oncologists recommend which chemotherapy regimen (if any) would be more effective for their patients.

To date, contemporary biomarkers, such as recurrence score, 70-gene MammaPrint risk score, risk of recurrence (ROR), and EndoPredict (EP), have not predicted benefit from the addition of a taxane or from dose-intense scheduling of chemotherapy for HR⁺ breast cancer (5–10). However, a recent analysis of estrogen receptor (ER)–positive cancers from the CALGB 9741 trial reported that a low sensitivity to endocrine therapy (SET_ER/PR) index (below 0.75) predicted survival benefit from dose-dense (2-weekly) versus conventional 3-weekly dosing schedules of the same anthracycline and paclitaxel treatments (11). On the other hand, the SET_ER/PR index was not associated with benefit from conventional dosing of anthracycline treatments in the SWOG 8814 trial, suggesting that prediction in CALGB 9741 might be related to paclitaxel or dose density in general (11, 12). The 28-gene SET_ER/PR index is a transcriptional assay comprising 18 target genes and 10 reference genes. It specifically measures ER) and progesterone receptor (PR) signaling activity, and notably, its results do not correlate with tumor proliferation (13). This assay was customized from earlier microarray-based research studies for use with formalin-fixed, paraffin-embedded (FFPE) tumor tissues and yields reproducible results within and between different laboratories (14–16). To our knowledge, the SET_ER/PR index is the first genomic assay to predict benefit from taxane-based chemotherapy or from a dose-dense regimen. One feasible scientific rationale is that endocrine activity is an important characteristic of the cellular differentiation that protects breast cells from cytotoxicity. For example, increased endocrine activity in breast cancer cells might prevent paclitaxel from stabilizing microtubules and/or support cellular recovery between treatments, which would diminish the therapeutic benefit from more frequent scheduling of chemotherapy treatments (11, 17–19). Beyond microtubule stabilization, taxanes have been shown to induce immunogenic cell death and modulate cell-cycle checkpoints (20). Increased endocrine activity in breast cancer cells might interfere with these pathways and/or support cellular recovery between treatments, thereby diminishing the therapeutic benefit from more frequent chemotherapy scheduling.

In this prospective-retrospective blinded clinical validation study, we tested the hypothesis that the SET_ER/PR index below 0.75 predicts benefit from weekly paclitaxel in patients with lymph node–positive, HR⁺/HER2− breast cancer from the seminal phase III GEICAM/9906 trial that evaluated the addition of weekly paclitaxel to anthracycline-based chemotherapy (21).

Materials and Methods

Study cohort

GEICAM/9906 (ClinicalTrials.gov identifier: NCT00129922) is a multicenter, open-label, randomized phase III trial comparing two adjuvant chemotherapy regimens after breast cancer surgery in women with lymph node–positive disease (21). A total of 1,246 patients with early breast cancer, aged 23 to 76 years, were enrolled. Participants were randomized to receive either six 21-day cycles of 5-fluorouracil, epirubicin, and cyclophosphamide (FEC; control arm) or four 21-day cycles of FEC followed by 8-weekly courses of paclitaxel (FEC-P; experimental arm). Patients with HR⁺ cancer received hormone treatment after chemotherapy. The median follow-up duration for the whole cohort was 10.3 years. Details of the study design and the patients’ characteristics have been reported previously. Patients whose tumors were confirmed by a central immunohistochemical (IHC) pathologic assessment to be ER and/or PR–positive and HER2-negative (i.e., HR⁺/HER2⁻) were eligible for this study (Supplementary Table S1; ref. 7). Molecular data from our published analysis of EP risk score (EPclin) were also included (22).

The GEICAM/9906 trial was performed in accordance with the Declaration of Helsinki and was approved by the ethics committees at all participating institutions and the Spanish Health Authority. All patients provided their written informed consent for therapy randomization and molecular analyses. The subsequent use of patients’ specimens without disclosure of patient identifiers met the waiver of informed consent policy criteria in accordance with the Declaration of Helsinki ethical guidelines. FFPE blocks from each patient’s primary tumor were collected at the time of surgery. Total RNA had been prepared from these tumor samples for previous studies (6, 7), and the residual RNA was used for this study. Full details on the GEICAM/9906 trial treatment protocols have been reported (21).

Study design and endpoints

The current study follows a formal prospective–retrospective design per REMARK criteria and per the guidelines for the use of archived clinical trial specimens for predictive biomarker evaluation in clinical trials (23, 24). The MD Anderson Cancer Center (MDACC) laboratory received RNA aliquots under an approved material transfer agreement and with institutional ethics approval. Laboratory personnel remained blinded to sample identity, subject’s treatment arm, and clinical outcome. An analysis plan was prespecified and conducted independently by the GEICAM/9906 statistician. The primary hypothesis was that an SET_ER/PR index below 0.75 in a subject’s breast cancer would predict benefit from the FEC + P treatment arm. The primary endpoint of distant recurrence–free interval (DRFI) was decided in advance and is defined as the time from the date of random assignment to the first date of diagnosis of distant recurrence or death attributable to breast cancer (25). Secondary endpoints were overall survival (OS), defined as the time from randomization to the date of death from any cause, and breast cancer–specific survival (BCSS), defined as the time from randomization to the date of death from breast cancer.

RNA processing and SET_ER/PR index determination

Total RNA from GEICAM/9906 cohort patients’ FFPE tumor tissue was isolated using the RNeasy Mini Kit (QIAGEN) and stored at −80°C. RNA purity and integrity were assessed both by spectrophotometry (NanoDrop Spectrophotometer 2000c, Thermo Fisher Scientific) and electrophoresis (2100 Bioanalyzer, Agilent Technologies). SET_ER/PR gene expression was measured using a multiplex hybridization technique (QuantiGene Plex assay, Thermo Fisher), as previously described (14, 15).

Statistical analysis

The prespecified approved statistical analysis plan was independently executed by the GEICAM/9906 statistician, testing the predictive capacity of the SET_ER/PR index. The SET_ER/PR index was evaluated as a continuous index (per unit) and a categorized index using the predefined cut point of 0.75 (11). Additional exploratory cut points were <1.00, the median value (high if above median), and quartiles of the SET_ER/PR index. χ² and Mann–Whitney tests were performed for categorical and continuous variable distributions, respectively.

Kaplan–Meier curves were used to estimate and compare the survival distribution across groups defined by categorical variables. The log-rank test assessed statistical significance, with its P value reported. The association of the continuous and categorized variables on the primary and secondary endpoints with treatment arm was evaluated using uni- and multivariable Cox proportional hazards regression analysis. In all the Cox regression models, calculated in the whole study cohort, the hazard ratios (HR) were computed along with a 95% confidence interval (95% CI), and the P value was reported. Multivariable analysis was adjusted for menopausal status and nodal status. Interaction tests of heterogeneity that assess the associations of variable categories with clinical outcomes between treatment arms were used. Bivariate Cox regression analyses were performed to evaluate the contribution of SET_ER/PR relative to the EPclin in predicting clinical outcomes and the trial’s treatment benefit. The proportional hazards assumption was assessed using Schoenfeld residuals and by visually inspecting log(−log) survival plots for all fitted Cox models.

All statistical tests were conducted at the two-sided 0.05 level of significance using R v4.1.1 software.

Results

Central biomarker assessment of the 1,045 tumor samples collected from patients in the GEICAM/9906 trial (83.9%) identified 681 (65.2%) that were HR⁺/HER2^–, with tumor RNA available from 647 (95%), and 567 (87.6%) providing an evaluable SET_ER/PR index result for this translational study cohort (Fig. 1A). Among those 567 evaluable patients, 288 (50.8%) received FEC treatment, whereas 279 (49.2%) were assigned to the experimental FEC-P arm. Their DRFI outcomes are shown by treatment arm in Fig. 1B. The outcomes closely resemble those previously reported for the HR⁺ subset of the original trial cohort (21). A comparison between the patients included in this biomarker study (n = 567) and the remaining HR⁺ patients from the GEICAM/9906 trial (n = 449) showed that both groups were well balanced. Although a nominal statistical difference was observed in median age (50.4 vs. 49.2 years, P = 0.046), this marginal difference of 1.2 years is not clinically meaningful. No other significant differences were found in key prognostic variables, suggesting no major selection bias (Supplementary Table S2).

Figure 1.

GEICAM/9906 retrospective translational study cohort. A, REMARK diagram of samples from GEICAM/9906 trial patients evaluated in this translational study. *, Inclusion criteria summarized as women aged 18–75, after surgery for unilateral breast cancer, with ≥1 positive axillary lymph node on histologic examination. B, Kaplan–Meier plot of DRFI for the 567 patients evaluable in this study, according to treatment arm. Covariables included in the multivariable Cox analysis were menopausal status and nodal status. MDACC, MD Anderson Cancer Center; QC, quality control.

Clinicopathologic characteristics were well balanced between both groups (Table 1). PR-negative status was more frequent in the FEC treatment arm (16.3% vs. 8.3%, P = 0.006) in the central biomarker assessment (Table 1) but was not different in the results that were reported from the local sites (20.1% vs. 16.9%, P = 0.368). The distributions of the SET_ER/PR index were similar in each treatment arm (Table 1). We observed a significant association between SET_ER/PR categories and ER IHC staining levels (P < 0.0001). Specifically, the SET_ER/PR low subgroup was enriched for tumors with lower ER expression (0%–50% staining), whereas the SET_ER/PR high subgroup predominantly comprised tumors with 100% ER staining. Additionally, SET_ER/PR low tumors were more frequently high-grade (G3) and had higher Ki67 levels compared with the SET_ER/PR high subgroup. Notably, no significant imbalance was found in the distribution of EPclin risk scores between SET_ER/PR subgroups, suggesting that these biomarkers capture different biological features. Using the cut point that was predefined for the primary analysis (SET_ER/PR below 0.75), the SET_ER/PR index was low in 92 patients (16.2%) and high in 475 patients (83.8%). SET_ER/PR index was prognostic for DRFI (HR per unit, 0.59; 95% CI, 0.40–0.87; P = 0.007), OS (HR, 0.67; 95% CI, 0.45–1; P = 0.049), and BCSS (HR, 0.56; 95% CI, 0.36–0.86; P = 0.008). The bivariate Cox analysis, including both SET_ER/PR and EPclin, showed that the SET_ER/PR index did not retain independent prognostic significance, reflecting significant prognostic information provided by EPclin (Supplementary Table S3). There was a significant interaction between SET_ER/PR status and treatment arm for DRFI (p_interaction = 0.046). Patients with an SET_ER/PR score below 0.75 had a lower risk of distant recurrence in the FEC + P arm compared with the FEC arm (HR, 0.48; 95% CI, 0.24–1; P = 0.049) after adjusting for menopausal status and the number of involved lymph nodes (1–3 vs. ≥4; Fig. 2A). On the other hand, no difference in treatment efficacy was observed among patients with SET_ER/PR ≥0.75 in univariable analysis (HR, 1.02; 95% CI, 0.70–1.47; P = 0.931) and after adjustment for menopausal status and nodal burden (Fig. 2B).

Table 1.

Patient and baseline characteristics of the GEICAM/9906 translational study cohort according to treatment arm.

Characteristic	FEC arm (n = 288), n (%)		P value	FEC + P arm (n = 279), n (%)		P value
	Low SETER/PR (n = 48)	High SETER/PR (n = 240)		Low SETER/PR (n = 44)	High SETER/PR (n = 235)
Median age, years (range)	50.1 (33.5–72.8)	49.6 (27.9–75.7)	0.609	49.8 (23.1–73.1)	50.98 (25.4–76.2)	0.656
Menopausal status at diagnosis	​	​	0.338	​	​	1.000
Premenopausal	24 (50%)	141 (58.8%)	​	23 (52.3%)	122 (51.9%)	​
Postmenopausal	24 (50%)	99 (41.3%)	​	21 (47.7%)	113 (48.1%)	​
Tumor size	​	​	0.632	​	​	0.833
≤2 cm	18 (37.5%)	102 (42.5%)	​	20 (45.5%)	118 (50.2%)	​
>2 and ≤5 cm	28 (58.3%)	123 (51.3%)	​	22 (50%)	106 (45.1%)	​
> 5 cm	2 (4.2%)	15 (6.3%)	​	2 (4.5%)	11 (4.7%)	​
Nodal status	​	​	0.451	​	​	0.554
1–3 positive nodes	26 (54.2%)	147 (61.3%)	​	26 (59.1%)	153 (65.1%)	​
≥4 positive nodes	22 (45.8%)	93 (38.8%)	​	18 (40.9%)	82 (34.9%)	​
Histologic type	​	​	0.527	​	​	0.117
Infiltrating ductal carcinoma	42 (87.5%)	210 (87.5%)	​	37 (84.1%)	194 (82.6%)	​
Infiltrating lobular carcinoma	3 (6.3%)	22 (9.2%)	​	3 (6.8%)	33 (14%)	​
Other	3 (6.3%)	8 (3.3%)	​	4 (9.1%)	8 (3.4%)	​
Histologic grade	​	​	0.001	​	​	0.002
G1	2 (4.2%)	46 (19.2%)	​	5 (11.4%)	41 (17.5%)	​
G2	19 (39.6%)	119 (49.6%)	​	12 (27.3%)	110 (46.8%)	​
G3	24 (50%)	58 (24.2%)	​	24 (54.6%)	60 (25.5%)	​
GX	3 (6.3%)	17 (7.1%)	​	3 (6.8%)	24 (10.2%)	​
ER IHC staining (central review, %)	​	​	<0.0001	​	​	<0.0001
0%	3 (6.3%)	8 (3.3%)	​	10 (22.7%)	6 (2.6%)	​
1%–50%	15 (31.3%)	22 (9.2%)	​	7 (15.9%)	28 (11.9%)	​
51%–99%	24 (50%)	109 (45.4%)	​	15 (34.1%)	89 (37.9%)	​
100%	5 (10.4%)	91 (37.9%)	​	9 (20.5%)	101 (43%)	​
Unknown	1 (2.1%)	10 (4.2%)	​	3 (6.8%)	11 (4.7%)	​
ER IHC status (central review)	​	​	0.510	​	​	<0.0001
Negative	3 (6.3%)	8 (3.3%)	​	10 (22.7%)	6 (2.6%)	​
Positive	44 (91.7%)	222 (92.5%)	​	31 (70.5%)	218 (92.8%)	​
Unknown	1 (2.1%)	10 (4.2%)	​	3 (6.8%)	11 (4.7%)	​
PR IHC status (central review)	​	​	<0.0001	​	​	0.241
Negative	21 (43.8%)	26 (10.8%)	​	6 (13.6%)	17 (7.2%)	​
Positive	25 (52.1%)	204 (85%)	​	35 (79.6%)	207 (88.1%)	​
Unknown	2 (4.2%)	10 (4.2%)	​	3 (6.8%)	11 (4.7%)	​
Ki67 IHC	​	​	0.0002	​	​	0.097
≤20%	32 (66.67%)	204 (85%)	​	30 (68.2%)	189 (80.4%)	​
>20%	15 (31.25%)	23 (9.58%)	​	10 (22.7%)	29 (12.3%)	​
Unknown	1 (2.08%)	13 (5.42%)	​	4 (9.1%)	17 (7.2%)	​
EPclin risk score	​	​	0.168	​	​	0.956
EPclin low	1 (2.08%)	29 (12.08%)	​	3 (6.8%)	24 (10.2%)	​
EPclin high	28 (58.33%)	168 (70%)	​	26 (59.1%)	165 (70.2%)	​
Unknown	19 (39.58%)	43 (17.92%)	​	15 (34.1%)	46 (19.6%)	​

Figure 2.

Survival analyses of DRFI between SET_ER/PR index and treatment in the GEICAM/9906 translational study cohort. A, Kaplan–Meier plot of DRFI for the subset of patients with SET_ER/PR < 0.75 randomly assigned to FEC vs. FEC + P. B, Kaplan–Meier plot of DRFI for the subset of patients with SET_ER/PR ≥ 0.75 randomly assigned to FEC vs. FEC + P. C, Kaplan–Meier plot of DRFI for patients in the FEC treatment arm scored as SET_ER/PR < 0.75 or SET_ER/PR ≥ 0.75. D, Kaplan–Meier plot of DRFI for patients in the FEC + P treatment arm scored as SET_ER/PR < 0.75 or SET_ER/PR ≥ 0.75. Covariables included in the multivariable Cox model were menopausal status and nodal status.

Patients in the FEC arm had a higher risk of distant recurrence when the SET_ER/PR score was below 0.75 (HR, 0.44; 95% CI, 0.27–0.72; P = 0.001) after adjusting for menopausal status and nodal burden (Fig. 2C). There was no prognostic difference by SET_ER/PR status in patients’ categorization in the FEC + P arm (HR, 0.93; 95% CI, 0.49–1.78; P = 0.830) after adjusting for menopausal status and nodal burden (Fig. 2D).

The SET_ER/PR index was explored further within clinical subgroups of menopausal status, tumor size, and nodal burden (Supplementary Fig. S1). The benefit of adding weekly paclitaxel was consistently greater in the SET_ER/PR low subgroup across all clinical strata, with HRs for distant recurrence being lower in the SET_ER/PR low patients compared with the SET_ER/PR high patients in every category analyzed.

Statistical significance was not reached for the secondary endpoints of OS (P_interaction = 0.352) or BCSS (P_interaction = 0.211). With respect to OS, a nonsignificant prognostic trend was observed in the subjects with an SET_ER/PR score below 0.75 (Fig. 3A; HR, 0.61; 95% CI, 0.29–1.30; P = 0.201), whereas no difference between treatment arms was found in the subgroup with a score ≥0.75 (Fig. 3B; HR, 0.84; 95% CI, 0.58–1.22; P = 0.365). Similar results were obtained for BCSS, showing a nonsignificant trend in the SET_ER/PR score below 0.75 subgroup (Fig. 3C; HR, 0.52; 95% CI, 0.23–1.17; P = 0.115) and no difference in those patients with a score ≥0.75 (Fig. 3D; HR, 0.89; 95% CI, 0.58–1.36; P = 0.590).

Figure 3.

Survival analyses of OS and BCSS between SET_ER/PR index and treatment in the GEICAM/9906 translational study cohort. A, Kaplan–Meier plot of OS for the patients with SET_ER/PR < 0.75 randomly assigned to FEC vs. FEC + P. B, Kaplan–Meier plot of OS for the patients with SET_ER/PR ≥ 0.75 randomly assigned to FEC vs. FEC + P. C, Kaplan–Meier plot of BCSS for the patients with SET_ER/PR < 0.75 randomly assigned to FEC vs. FEC + P. D, Kaplan–Meier plot of BCSS for the patients with SET_ER/PR ≥ 0.75 randomly assigned to FEC vs. FEC + P. Covariables included in the multivariable Cox model were menopausal status and nodal status.

Analyses using other exploratory cut points for the SET_ER/PR index [1, 1.43 (median), and quartiles] did not demonstrate any significant interaction by treatment arm (Table 2). Similarly, the evaluation of DRFI HRs across treatment arms as a function of the SET_ER/PR index had broad CIs but may illustrate a nonlinear relationship in which the hazard function plot crosses below 1.0 (favoring the FEC + P arm) with lower SET_ER/PR scores (below 0.75; Fig. 4).

Table 2.

Exploratory analysis of other SET_ER/PR index cut points for prediction of DRFI benefit in the GEICAM/9906 translational study cohort.

Description of cut point	Cut point range of SETER/PR index	P interaction
Predefined	<0.75 vs. ≥0.75	0.046
Arbitrary	<1.00 vs. ≥1.00	0.740
Median	≤1.43 vs. >1.43	0.418
IQR	0.99–1.78	0.456
Quartiles	≤0.98; 0.99–1.43, 1.44–1.78, >1.78	0.405
None	Continuous	0.665

Abbreviation: P_interaction, P value for interaction between SET_ER/PR index and chemotherapy treatment arm on DRFI. Bold values indicate statistically significant results (P < 0.05).

Figure 4.

Hazard function plot of the rate of hazard for distant recurrence in the FEC + P treatment arm relative to the FEC treatment arm as a function of the SET_ER/PR index as a continuous variable. HR is plotted on the y-axis (log scale; values < 1 favor FEC + P) against the SET_ER/PR index on the x-axis (linear scale). Red and blue distributions correspond to the FEC and FEC + P treatment arms, respectively. Dashed lines indicate 95% CIs.

Discussion

This study independently confirmed in a second clinical trial that a tumoral SET_ER/PR index below 0.75 predicted benefit from adjuvant chemotherapy that includes dose-dense paclitaxel, that is, 2-weekly or lower-dose weekly treatments. There were some differences in the design of the GEICAM/9906 and CALGB 9741 clinical trials. In the current analysis from GEICAM/9906, weekly paclitaxel was added to anthracycline-based chemotherapy conventionally dosed every 21 days, and HR⁺/HER2^– breast cancer samples were analyzed. In the previous analysis from CALGB 9741, dose-dense (14-day) and conventional (21-day) dosing schedules of paclitaxel and anthracycline-based chemotherapy were compared, and HR⁺ breast cancer samples were analyzed (11). Despite these differences, the SET_ER/PR index remained predictive at its predefined cut point.

Generally, arguments supporting the superiority of dose-dense chemotherapy rely on the principles of low-volume Gompertzian growth kinetics after each treatment cycle. One principle is that the regrowth of low-volume disease has an initially higher rate of cellular proliferation, which would imply greater sensitivity to the next chemotherapy treatment dose. However, neither the rate of proliferation nor the molecular subtype of the original tumor has predicted benefit from dose-dense chemotherapy for HR⁺ breast cancer (6–8, 17, 26). Instead, our proposed rationale for this biomarker–treatment interaction is that cancers with low endocrine-related activity would have a higher cell death rate and an impaired ability of cancer cells to recover when there is a shortened interval between successive doses of chemotherapy (18). Interestingly, low SET_ER/PR scores did not predict benefit from the addition of conventionally dosed anthracycline-based chemotherapy to endocrine therapy in the S8814 trial (12). Taken together, these results suggest that low endocrine activity in the cancer predicts benefit from dose-dense paclitaxel treatments but does not fully exclude the possibility that it might also predict benefit from dose-dense anthracycline treatments (11).

We do note that the SET_ER/PR index includes the gene MAPT, which encodes the microtubule-associated tau protein. Tau protein is a known inhibitor of the paclitaxel binding site on β-tubulin, and high levels of MAPT gene expression or tau protein expression have been reported to be associated with a lower rate of pathologic complete response (pCR) from neoadjuvant taxane-containing chemotherapy. However, it is strongly associated with ER-positive status such that adjuvant endocrine therapy affects outcomes favorably for the high-tau subgroup (13, 19, 27). Hence, it is also conceivable that the SET_ER/PR index might demonstrate predictive ability that is specific to taxane-based regimens or potentially to paclitaxel alone. Our results also suggest that the observed interaction between low endocrine transcriptional activity (measured using the SET_ER/PR index) and benefit from weekly paclitaxel might follow a nonlinear relationship, with a pharmacologically relevant threshold around 0.75 units, rather than a continuous relationship through the full range of SET_ER/PR index values. This is supported by the lack of predictive interaction at other predefined cut points (Table 2) and the nonlinear hazard function plot (Fig. 4). Conceptually, there might be negligible benefit from dose-dense paclitaxel treatments if the endocrine activity in the cancer cells is sufficient to protect cells from lethal damage and facilitate recovery between doses.

Low endocrine transcriptional activity in breast cancer is associated with higher histologic grade, increased proliferation, and increased numbers of stromal tumor-infiltrating lymphocytes (TIL; refs. 28, 29). Interestingly, patients with HR⁺/HER2⁻ cancers with lower levels of TILs might benefit from the addition of docetaxel to anthracycline treatments (although there was no significant interaction), but that was different from the significant predictive interaction seen with triple-negative cancers (30). We plan to evaluate TILs in the GEICAM/9906 to determine whether the combination of TILs and the SET_ER/PR index might further refine the prediction of benefit from the addition of weekly paclitaxel.

A limitation of this study is that the sample size is small for a test of predictive interaction. Multivariable Cox analyses were adjusted for menopausal status and nodal status as these are currently the most clinically relevant confounders in relation to chemotherapy decision-making for HR⁺ breast cancer. Statistical power was further limited by the finding that only 16% of the cancers had an SET_ER/PR index below 0.75, fewer than the 41% observed in the former CALGB 9741 trial (11). A fundamental difference between these two studies is that GEICAM/9906 samples had undergone central pathology evaluation (whereas CALGB 9741 had not) and excluded subjects with known HER2⁺ cancer. Assuming that approximately 15% of HR⁺ cancers would also be HER2⁺, and with past observations that most HER2–positive cancers have an SET_ER/PR index below 0.75, we might have anticipated from the CALGB 9741 population that approximately 26% of HR⁺/HER2⁻ cancers would have an SET_ER/PR index below 0.75. The observed frequency in GEICAM/9906 (16%) is lower than anticipated from CALGB 9741, and although both enrolled pre- and postmenopausal women with node-positive cancer, it is conceivable that the GEICAM/9906 trial population more frequently had high tumoral endocrine activity. That is supported by the high rates of PR positivity and low proliferation (Ki67 < 20%) in this study cohort (Table 1). Further, our characterization of the GEICAM/9906 cohort demonstrates that although the SET_ER/PR index correlates with ER expression by IHC, it provides independent predictive value (Table 1). The index identifies patients with low endocrine transcriptional activity—even within the high-IHC expression group—who can derive substantial benefit from weekly paclitaxel. This predictive capacity was not captured by EPclin in this cohort, underlining that SET_ER/PR measures a distinct biological dimension linked to chemosensitivity rather than just metastatic potential.

Another limitation was the imbalance of PR⁻ tumors at central review (less frequent in the FEC + P arm) although the distribution of the SET_ER/PR index remained similar (Table 1). Both treatment arms in GEICAM/9906 included 5-fluorouracil, which is no longer included. However, we doubt this influenced our findings because 5-fluorouracil was not used in the prior CALGB 9741 trial, in which the SET_ER/PR index had predicted benefit from the dose-dense chemotherapy (11). Furthermore, although patients in the FEC + P arm received only 8 weeks of paclitaxel, the dose was 100 mg/m², resulting in a high dose intensity (cumulative dose of 800 mg/m²) that is close to the 960 mg/m² delivered in modern 12-week schedules at 80 mg/m². We acknowledge that a small sample size and a P value of 0.046 for the interaction analysis could imply a risk of a false-positive finding. However, it is critical to note that this study is not an exploratory analysis but a prospective–retrospective validation of a prespecified cut point (0.75) established in the independent CALGB 9741 trial. The consistency of this threshold across different randomized cohorts reinforces that our finding reflects a robust biological interaction rather than a chance occurrence.

We recognize that GEICAM/9906 began enrollment about 25 years ago and the use of chemotherapy for node-positive HR⁺/HER2⁻ cancer has since become focused on a higher-risk subset of patients (with a higher proportion likely to have SET_ER/PR index below 0.75). Nevertheless, weekly paclitaxel is still a standard adjuvant chemotherapy regimen for HR⁺/HER2⁻ cancer, and so the findings from this study are applicable to contemporary clinical practice.

Large randomized trials have demonstrated that some patients with HR⁺/HER2⁻ breast cancer do not benefit from any chemotherapy (4, 5). However, the Oxford overview analyses have also demonstrated that improvements in adjuvant chemotherapy regimens, such as the inclusion of taxanes and dosing frequency, also translate to patients with HR⁺/HER2− breast cancer (1–3). Therefore, we must be cautious about extrapolating predictive biomarker findings related to early generations of adjuvant chemotherapy to more contemporary chemotherapy regimens (4, 5, 31, 32). For example, biomarkers that predicted benefit from an older chemotherapy regimen (vs. no chemotherapy) did not predict who benefited from a more contemporary regimen that demonstrated superiority over the older regimen (4–8). Similarly, we must be cautious that predictive signals might actually come from HR⁻ or HER2⁺ subtypes (33, 34). Finally, it is difficult to accurately infer chemo prediction when there is an imbalance between treatment arms or when compared across different clinical trials or treatment subgroups selected from registry studies because prognostic differences might still elude statistical efforts to propensity match for risk factors (35, 36). Therefore, it is probably too simplistic to assume that prediction to chemotherapy by a biomarker would apply to any regimen even though that is commonly assumed. Rather, one also has to consider that significant pharmacologic differences between chemotherapy regimens might require distinct predictive biomarkers.

These findings have significant implications for neoadjuvant treatment for high-risk HR⁺/HER2⁻ breast cancer. Recent landmark trials, such as KEYNOTE-756 and CheckMate 7FL (28, 37), have demonstrated that adding immunotherapy to neoadjuvant chemotherapy increases pCR rates in high-risk HR⁺/HER2⁻ breast cancer. This synergy from chemotherapy and immune activation might be enhanced if the most effective chemotherapy regimen for a patient’s breast cancer can be selected to maximally stimulate the immune response.

GEICAM/9906 is the second randomized clinical trial from which an independent prospective–retrospective analysis demonstrated significant prediction of benefit from adjuvant chemotherapy containing paclitaxel when the SET_ER/PR index in a primary HR⁺/HER2⁻ breast cancer is below 0.75. This establishes the clinical utility of the SET_ER/PR index as the first biomarker to predict benefit from contemporary taxane-based chemotherapy and the first biomarker to predict benefit between recent adjuvant chemotherapy regimens for HR⁺/HER2− breast cancer. Of course, this exciting new utility prompts important new questions, such as whether the biomarker also predicts benefit from docetaxel or dose-dense administration of anthracyclines or whether weekly paclitaxel alone would be sufficient for some patients. We look forward to assiduously addressing such questions in future studies to further refine the predictive utility of this biomarker.

Supplementary Material

Supplementary Figure S1

Forest plot for the GEICAM/9906 translational study cohort primary endpoint of DRFI, on the FEC vs FEC + P arms for SETER/PR cutpoint “0.75” categories.

Supplementary Table S1

Representativeness of Study Participants.

Supplementary Table S2

Patient and baseline characteristics of GEICAM/9906 hormone receptor positive (HR+) breast cancer cohort.

Supplementary Table S3

Bivariable Cox regression analysis of DRFI between SETER/PR index and EPclin risk score in the GEICAM/9906 translational study cohort.

Data Availability

To protect participant privacy and comply with institutional data protection policies, the datasets analyzed during the current study, including deidentified clinical outcomes and SET_ER/PR index values, are not publicly available. Nevertheless, these datasets are available from the GEICAM Translational Research Program upon reasonable request. Access is contingent upon formal application and approval of a research protocol by the scientific committee. Correspondence about data access should be sent to secretaria-cientifica@geicam.org.

Authors’ Disclosures

M. Martín reports personal fees from Amgen, AstraZeneca, Daiichi Sankyo, Eli Lilly, Pfizer, PharmaMar, and Taiho Oncology, as well as grants and personal fees from Novartis, Puma Biotechnology, and Roche/Genentech outside the submitted work. A. Rodriguez-Lescure reports grants from GEICAM during the conduct of the study, as well as personal fees from Lilly and grants and personal fees from Roche, Pfizer, Novartis, Gilead, and AstraZeneca outside the submitted work. N. Martinez-Jañez reports nonfinancial support from Pfizer and personal fees from AstraZeneca, Novartis, Lilly, and Daiichi Sankyo outside the submitted work. E. Alba reports grants from Pfizer and AstraZeneca and personal fees from Exact Sciences, Pfizer, Novartis, Eli Lilly, and Bristol Myers Squibb outside the submitted work. I. Álvarez reports other support from Roche, Pfizer, Gilead, and AstraZeneca/Daiichi Sankyo, as well as grants from Roche, Pfizer, MSD, Gilead, Novartis, and AstraZeneca/Daiichi Sankyo outside the submitted work. M.A. Seguí reports grants from GEICAM during the conduct of the study. V. Valero reports other support from Genomic Health outside the submitted work. J.J. Ponce-Lorenzo reports personal fees from Novartis, AstraZeneca/Daiichi Sankyo, Lilly, MSD, Pfizer, Roche, Pierre Fabre, and Gilead outside the submitted work. A. Guerrero-Zotano reports personal fees from Roche, Gilead, Pierre Fabre, Lilly, AstraZeneca, and Menarini Stemline outside the submitted work. F. Rojo reports grants and personal fees from Roche, AstraZeneca, Menarini Stemline, Astellas, AbbVie, Sophia Genetics, and Agilent, as well as personal fees from Bristol Myers Squibb, Merck, MSD, Lilly, Novartis, GSK, and BeiOne outside the submitted work. W.F. Symmans reports grants from the Breast Cancer Research Foundation and the Cancer Prevention and Research Institute of Texas during the conduct of the study, as well as other support from Delphi Diagnostics and IONIS Pharmaceuticals and personal fees from the Miami Breast Cancer Conference, AstraZeneca, the Southwest Oncology Group, Memorial Sloan Kettering, the New York Metropolitan Breast Group, the St. Gallen Breast Cancer Conference, and Saga Diagnostics outside the submitted work. W.F. Symmans also reports a patent for US Patent # 11,459,617 issued and licensed to Delphi Diagnostics and a patent for Patent Application UTSC.P1519US.P1 pending and licensed to Delphi Diagnostics, as well as stock ownership in EIGER Biopharmaceuticals (publicly traded). No disclosures were reported by the other authors.

Authors’ Contributions

M. Martín: Conceptualization, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing. A. Rodriguez-Lescure: Investigation, writing–review and editing. C. Reboredo: Investigation, writing–review and editing. E. Chen: Data curation, validation, investigation, writing–review and editing. M. Ruiz-Borrego: Investigation, writing–review and editing. A. Santaballa Bertran: Investigation, writing–review and editing. C.A. Rodríguez: Investigation, writing–review and editing. N. Martinez-Jañez: Investigation, writing–review and editing. E. Alba: Investigation, writing–review and editing. K. Tran: Data curation, validation, investigation, writing–review and editing. I. Pelaez Fernandez: Investigation, writing–review and editing. I. Álvarez: Investigation, writing–review and editing. M.A. Seguí: Investigation, writing–review and editing. A. de la Cruz: Investigation, writing–review and editing. V. Valero: Data curation, formal analysis, investigation, writing–review and editing. A. Antón-Torres: Investigation, writing–review and editing. R. Andrés: Investigation, writing–review and editing. K. Amillano: Investigation, writing–review and editing. J.J. Ponce-Lorenzo: Investigation, writing–review and editing. S. Dominguez Fernandez: Investigation, writing–review and editing. J. Herranz: Data curation, formal analysis, investigation, writing–review and editing. R. Rincon: Data curation, formal analysis, investigation, writing–review and editing. R. Caballero: Conceptualization, formal analysis, supervision, investigation, methodology, writing–review and editing. A. Guerrero-Zotano: Investigation, writing–review and editing. F. Rojo: Conceptualization, formal analysis, supervision, investigation, methodology, writing–review and editing. W.F. Symmans: Conceptualization, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing.